Daily AI CatchupAI Catchup — Tuesday, August 25, 2026
Tuesday, August 25, 2026Inference
NVIDIA's Groq 3 LPX Enters Full Production, Delivering 4x Faster AI Inference for Agentic Tasks
NVIDIA's new inference chip completes agentic AI tasks in minutes instead of hours with record token generation speeds.
MysteryMystery AI Model 'Ox Alpha' Breaks OpenRouter Records With 26 Trillion Tokens in 4 Days
An anonymous AI model hit 327K concurrent users and processed 26T tokens—but nobody knows who built it.
SecurityLLMs Can Exploit Inference Engine Vulnerabilities to Control Host Machines, Security Research Warns
AI models can run token sequences that hijack GPU loading software, giving them access to datacenter networks.
Video-GenerationAlibaba Launches Wan 3.0 Video Model, Generating 30-Second Videos From Text Prompts
Alibaba's new generative video model can create half-minute videos from text, upping competition in AI video generation.
NvidiaNVIDIA Extends CUDA Support to RISC-V Architecture, Opening New Hardware Paths for GPU Compute
NVIDIA is bringing CUDA to RISC-V CPUs, letting non-x86 processors feed GPU clusters for the first time.